The Weighted Moving Average method is usually used for smoothing purposes. However, it can be used to forecast $Y(t+1)$ based on the last n observed data.

In real-world problems, forecasting in very short horizon $(h = 1)$ is not very interesting.

I would like to know if it is possible to use larger forecasting horizons in WMA method like in HoltWinter and ARIMA techniques ?

  • $\begingroup$ Is it possible?: Sure, why not? The WMA is a recursive formula, so to get an n-step ahead forecast, plug in the previous (n-1)-step ahead forecasts to the right hand side. You can can choose some data or simulate it. See how well the WMA forecast performs for multiple steps out of sample and compare it to other techniques like ARIMA and HW. $\endgroup$ Commented Aug 1, 2015 at 12:15
  • $\begingroup$ But calculating $Y(t+h)$ at observation $t$, requires the calculation of $Y(t+h-1), Y(t+h-2), ..., Y(t+h-n)$. That is the weighted average of the last $n$ observed data, which is not available at observation $t$. $\endgroup$
    – mathematix
    Commented Aug 1, 2015 at 22:03

1 Answer 1


For an exponentially WMA you would get a flat forecast. My answer is derived from the information at http://people.duke.edu/~rnau/411avg.htm . Say you are using an exponentially Weighted Moving average for a series $\{Y_t\}$. You can write it as $$ S_{t}= \alpha Y_{t}+(1-\alpha)S_{t-1} $$ Your smoothed value $S_t$ is the point forecast for the next time period, i.e. $\hat Y_{t+1}=S_t$. Thus $$ \hat Y_{t+1}= \alpha Y_{t}+(1-\alpha)\hat Y_{t} $$

The only thing that you can do if you wanted to forecast 2 periods into the future is substitute in $\hat Y_{t+1}$ for $Y_{t+1}$ $$ \hat Y_{t+2}= \alpha \hat Y_{t+1}+(1-\alpha)\hat Y_{t+1}=\hat Y_{t+1} $$

and so $\hat Y_{t+1}=\hat Y_{t+2}=\hat Y_{t+3}=\hat Y_{t+4}=...$ and so on.

It is not that you can't forecast farther into the future, rather the forecast is flat. A flat forecast is not necessarily a bad forecast either. The random walk without drift is a flat forecast, and for certain time series, like crude oil for example, it outperforms most all other time series models some of which are very sophisticated.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.